corexense-mcp
CoreXense-MCP is a non-actuating MCP evidence gateway for synthetic physical-AI validation, decision traces, and replay audits.
List supported synthetic fruit decision scenarios and their explicit limits.
Run bounded local evidence qualification for a scenario case and return a decision plus trace.
Explain delayed or conflicting synthetic evidence using TWS-Belief reconstruction.
Review synthetic fruit episodes with cloud use controlled by host configuration.
Audit explicit FIRA simulation replay bundles from report and trace inputs without authorizing execution.
It exposes no actuator or arbitrary filesystem tools; live semantic review requires external cloud access, and offline results are explicit mocks.
Integrates with NVIDIA Nemotron models, hosted via Nebius Token Factory, to list accessible model IDs and perform bounded semantic evidence reviews with host-controlled cloud access.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@corexense-mcp review episode 001 and trace why the workflow paused"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
CoreXense-N: An MCP Evidence Gateway for Physical AI
Validate evidence, request bounded re-observation, and trace why a workflow pauses. The submission is a software review prototype. It does not control robot motors.
Run
Use Python 3.11+:
git clone https://github.com/xli431/corexense-n-hackathon.git
cd corexense-n-hackathon
python -m venv .venv
source .venv/bin/activate
python -m pip install -e '.[test]'
python -m pytest -q
python tools/run_mcp_sprint.py runs/offline-001Open runs/offline-001/review.html. The runner uses actual MCP stdio communication.
Default semantic reviews are explicit mocks; offline success is not a live-model result.
Related MCP server: Ackrite
NVIDIA and Nebius
First list the NVIDIA Nemotron models accessible to your account, with no inference:
python tools/run_mcp_sprint.py runs/catalog --live --prompt-keyThen pass one exact returned model ID:
python tools/run_mcp_sprint.py runs/live-001 --live --prompt-key --model EXACT_MODEL_IDCredentials stay in a hidden local prompt. This walkthrough makes at most two inference calls unless an optional simulation episode is supplied (then three). There is no automatic retry. Nebius Token Factory hosts NVIDIA Nemotron semantic review. Local checks and elapsed evidence validity remain authoritative.
The cloud response may correctly result in HOLD after the 250 ms synthetic evidence window expires. That is not robot performance, a task failure rate, or a safety proof.
Recorded live check: three requests, 2,674 reported tokens. The recovery fixture
received model ALLOW but remained HOLD because its evidence expired during the
audit. Another response was truncated at the default 256-token output budget.
Use --audit-max-tokens 1024 for a separately recorded budget follow-up; the
freshness policy and strict response validation remain unchanged.
Build a verified summary from a completed run (no API requests):
python tools/build_evidence_report.py runs/live-001/mcp_report.json runs/live-summary-001MCP tools
list_scenarios: list synthetic fixtures.review_scenario: original local fruit qualification and trace.review_episode: synthetic episode-bound semantic review and post-audit check.explain_belief: synthetic delayed/conflicting evidence reconstruction.review_simulation_episode: explicit FIRA JSON replay audit, always non-executable.
The server offers no actuator or arbitrary filesystem tools. Host configuration
controls cloud access. corexense-mcp starts the stdio service after installation.
FIRA prior work and current integration
The existing MuJoCo/RoboSim produce scene and its UI remain separate work.
An owner may supply a result.json / trace.jsonl episode directory using
--simulation-episode /absolute/path/to/episode. Its integrity is checked and
the same source hash is bound to the semantic review. This is historical auditing,
not closed-loop inference controlling the simulated or physical robot.
Evaluation and limitations
python tools/run_robustness.py runs/robustness-001This runs 1,920 synthetic policy evaluations and 17 injected transport contracts. These are fixture diagnostics, not independent physical trials or live model accuracy. The hash ledger detects changes relative to a retained tip; it does not authenticate the original source. Real fruit input remains blocked until a sensor adapter is commissioned. No supervised physical controller is included.
Model tests
Two recorded 1,024-token Nano walkthroughs produced six complete, schema-valid responses. Healthy and recovery model ALLOW verdicts still resulted in HOLD after the synthetic evidence window expired. See model test results for the 256/1,024-token comparison, run IDs, and a separately labeled author-reported five-model summary. These are small fixture checks, not a general model ranking or robot evaluation.
Source boundaries and prior work
The core derives from the existing CoreSense/CoreXense-N work; preserve the Apache license and notices. TWS-Belief is vendored under MIT with a pinned source manifest. Private operation_logging, review, perception, raw media, hardware calibration, FIRA upstream UI, and PEG research are excluded. No third-party dataset is bundled. The upstream TWS-Belief CLI/dataset adapters are not exposed by this demo; only the reconstruction API used by the MCP tool is qualified in this package.
See docs/REUSE_MAP.md for related projects. New sprint work: bound MCP episode review, simulation replay adapter, TWS-Belief tool, end-to-end client and reports. Live results, physical observations and future capabilities must be reported separately.
FIRA background acknowledgments: Voxel51 FiftyOne for data inspection, SKAI Intelligent for simulation support, and HawkRobo for earlier FIRA sponsorship. These credits do not imply a new partnership or endorsement of this submission.
Offline reproduction remains available after cloud credit expiry. New live inference requires valid Nebius account access.
See docs/SUBMISSION_STORY.md for the submission draft and
docs/SPRINT_CLOSEOUT.md for verified results and remaining steps. SKAI course
assets remain outside this repository. Local inspection opened the USD stage with
no unresolved dependencies; simulator execution and training remain separate work.
For reproduction after cloud credits expire, see the judge runbook. The source export has been independently installed and verified with 103 tests, an installed MCP server launched outside the source tree, and the offline fault matrix.
Available Tools
5 toolsexplain_beliefC
Explain delayed or conflicting synthetic evidence with TWS-Belief.
Inputs: case (required).
list_scenariosA
List supported synthetic fruit decision scenarios and explicit limits.
Inputs: no named parameters.
review_episodeC
Review a synthetic fruit episode; cloud use is controlled by the host configuration.
Inputs: case (required).
review_scenarioC
Run bounded local evidence qualification, returning a decision and trace.
Inputs: case (required).
review_simulation_episodeC
Audit an explicit FIRA simulation replay bundle; never authorize execution.
Inputs: trace (required), report (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 5 tools
Three of the five tools (review_scenario, review_episode, review_simulation_episode) share the same 'review_*' verb and closely related semantics; descriptions differentiate them (local qualification vs. synthetic episode vs. FIRA replay audit), but the scenario-vs-episode boundary remains fuzzy and invites misselection. list_scenarios and explain_belief are clearly distinct.
All five names follow a consistent snake_case verb_noun pattern (list_scenarios, review_scenario, explain_belief, review_episode, review_simulation_episode). No mixed conventions or casing deviations.
Five tools is well within the sweet spot for a narrowly scoped decision-review server. Each tool maps to a distinct capability (enumeration, review, explanation, episode audit, simulation audit) and none appear redundant padding.
The surface covers enumeration, review, explanation, and replay auditing, but there is no tool to fetch or inspect a specific scenario/episode by identifier, nor to list available episodes or query belief state directly. These are notable gaps that could force agents into dead ends after listing.
Maintenance
Related MCP Connectors
Independent AI-agent reviews: trust checks, evidence scorecards, incident registry, recommendations.
Physical-world evidence and operability checks with provenance and explicit data gaps.
Pre-execution governance for AI agents. Deterministic PASS/FAIL/REVIEW verdicts, replayable proof.
Authority, mandate, and policy checks for AI finance with replayable evidence. Not live execution.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceEnables step-debugging, deterministic replay, and signed audit evidence for AI agents, compliant with EU AI Act.79 PyPIMIT
- AlicenseBqualityDmaintenanceEnables AI agents to verify technical claims against supplied evidence, identify unsupported assumptions and contradictions, and recommend the smallest next check before acting.5MIT
- AlicenseNot gradedqualityBmaintenanceEnables deterministic verification of AI agent decisions and actions, providing PASS/FAIL/ABSTAIN verdicts with replayable proofs and an optional signed receipt ledger.Apache 2.0
- AlicenseAqualityBmaintenanceLets an assistant read recorded production AI agent runs — steps, model and tool calls with arguments, and outcomes — along with the findings of comparison runs that checked a prompt or model change against those recordings, and verify a signed evidence bundle offline with no account or network. It is strictly read-only: no tool starts a replay or comparison, so nothing can spend money or act without a person deciding.5Apache 2.0